A Robust Affine Invariant Point Extraction Algorithm for Image Registration

Affine invariance is a critical property for control points in point based image registration. In this paper, a highly robust affine invariant point extraction algorithm is proposed. For a pair of two shapes, they are handled as the longitudinal segments parallel to the line connecting centroids of two shapes, and then some longitudinal segments with the uniformly sampled interception on y-axis are adopted to calculate a descriptor which reflects the relative attitude between two shapes. In the meantime, the intersection points of sampled longitudinal segments and the contours of the shapes are taken as control points. After the corresponding shapes are found based on minimal distance between descriptors, the affine invariant control points from corresponding shapes are then used to estimate the transformation between images. Experimental results on synthetic and real data show that, our algorithm outperforms SC with higher precision.